1 Processamento dos dados

source("./load_files.R")
data_files_selected <- 
  list.files(
    pattern = ".*thais_.*csv$",
    recursive = TRUE,
    ignore.case = TRUE
    )

data_files_and_size <- 
  sapply(data_files_selected, file.size)

files_to_include_in_dataframe <- 
  tibble(
    "Files" = names(data_files_and_size),
    "Size (in MB)" = data_files_and_size/1E6
   )

skimr::skim(files_to_include_in_dataframe)
Data summary
Name files_to_include_in_dataf…
Number of rows 24
Number of columns 2
_______________________
Column type frequency:
character 1
numeric 1
________________________
Group variables None

Variable type: character

skim_variable n_missing complete_rate min max empty n_unique whitespace
Files 0 1 128 130 0 24 0

Variable type: numeric

skim_variable n_missing complete_rate mean sd p0 p25 p50 p75 p100 hist
Size (in MB) 0 1 2.18 3.48 0.32 0.38 0.6 2.1 10.99 ▇▁▁▁▁ 
cdr <- load_cdr(names(data_files_and_size))

cdr %<>% 
  mutate(
    expgroup = case_when(
      str_detect(file, "thaisnovoheader") ~ "nh",
      str_detect(file, "thais_29") ~ "29",
      str_detect(file, "thais_66") ~ "66",
      TRUE ~ "unknown"),
    cycle = case_when(
      str_detect(file, "R0_R2") ~ "R0_R2",
      str_detect(file, "R0_R3") ~ "R0_R3",
      str_detect(file, "R0_R4") ~ "R0_R4",
      str_detect(file, "R2_R3") ~ "R2_R3",
      str_detect(file, "R2_R4") ~ "R2_R4",
      str_detect(file, "R3_R4") ~ "R3_R4",
      TRUE ~ "unknown"),
    time = case_when(
        str_detect(file, "Initial") ~ "initial",
        str_detect(file, "Final") ~ "final",
        TRUE ~ "unknown")) %>% 
  select(cdr3, cycle, time, expgroup, everything())

cdr %<>% 
  group_by(cdr3, expgroup, cycle) %>% 
  arrange(desc(time), .by_group = TRUE) %>% 
  mutate(
    fcp = cdrp / lag(cdrp, default = first(cdrp)),
    fcq = quantity / lag(quantity, default = first(quantity))
  ) %>% 
  select(cdr3:quantity, fcp, fcq, everything())
cdr %>% 
  filter(time == "final") %>% 
  # filter(str_detect(cycle, "R0")) %>% 
  group_by(expgroup, cycle, time) %>% 
  arrange(desc(fcp)) %>% 
  slice_head(prop = .1) %>% 
  # slice_head(n = 1000) %>% 
  ggplot(aes(expgroup, log10(fcp))) +
    geom_violin(aes(fill = expgroup, color = expgroup), alpha = 0.5) +
    geom_jitter(aes(shape = expgroup), alpha = 0.6, size = 1) +
    stat_summary(
      fun = mean,
      fun.min = mean,
      fun.max = mean,
      geom = "crossbar",
      # width = 0.5,
      aes(color = expgroup)
    ) +
    facet_grid(. ~ cycle)

# cdr %>% 
#   filter(str_detect(cycle, "R0")) %>% 
#   filter(time == "final") %>% 
#   group_by(expgroup, cycle, time) %>% 
#   arrange(desc(fcp)) %>% 
#   slice_head(n = 1000) %>% 
#   ggplot(aes(fcp, color = expgroup, fill = expgroup)) +
#     geom_density(stat = "bin", alpha = 0.3) +
#     facet_grid(cycle ~ expgroup)

cdr %<>% 
  group_by(expgroup, cycle, time) %>% 
  arrange(desc(fcp)) %>% 
  slice_head(prop = .1) %>% 
  mutate(
    threshold = mean(log10(fcp))
    ) %>% 
  mutate(
    rich = if_else(
        (log10(fcp) >= threshold) &
          # (time == "final") &
          # (str_detect(cycle, "R0")),
          (time == "final"),
      "rich",
      "medium")) %>% 
  full_join(cdr) %>% 
  mutate(
    rich = if_else(
      is.na(rich),
      "poor",
      rich)) %>% 
  mutate(
    rich = factor(rich,
                  levels = c("rich", "medium", "poor"))
    ) %>% 
  mutate(
    threshold = if_else(
      is.na(threshold),
      0,
      threshold
    )
  )

cdr %>% 
  filter(rich == "rich") %>%
  ggplot() +
    geom_violin(aes(expgroup, log10(fcp), fill = expgroup)) +
    geom_jitter(aes(expgroup, log10(fcp), shape = expgroup), alpha = .2) +
    facet_grid(rich ~ cycle)

cdr %>% 
  filter(!rich == "rich") %>%
  ggplot() +
    geom_violin(aes(expgroup, log10(fcp), fill = expgroup)) +
    geom_jitter(aes(expgroup, log10(fcp), shape = expgroup), alpha = .2) +
    facet_grid(rich ~ cycle)

2 Tentativa de Clusterização

2.1 Clusterização Experimento 29

library(Rtsne)
set.seed(42)

tsne_df <- cdr %>% 
  # filter(!rich == "poor") %>%
  filter(time == "final") %>% 
  filter(cycle == "R0_R4") %>% 
  filter(expgroup == "29")

tsne_df %>% 
  summarise(across(everything(), ~ all(sum(is.na(.x))))) %>% 
  select(where(isTRUE))
## # A tibble: 1 x 2
## # Groups:   expgroup, cycle [1]
##   expgroup cycle
##   <chr>    <chr>
## 1 29       R0_R4
tsne_df %>% 
  filter(is.na(threshold)) %>% 
  select(rich, threshold, quantity)
## # A tibble: 0 x 6
## # Groups:   expgroup, cycle, time [0]
## # … with 6 variables: expgroup <chr>, cycle <chr>, time <chr>, rich <fct>,
## #   threshold <dbl>, quantity <int>
set.seed(42)
tsne_out <- 
  tsne_df %>% 
    ungroup() %>% 
    select(!where(is.character)) %>% 
    select(!c(which(apply(., 2, var)==0))) %>% 
    unique() %>% 
    Rtsne(
      X = .,
      dims = 3,
      perplexity = 420,
      theta = 0.1,
      max_iter = 2E3,
      verbose = T,
      pca_center = T,
      pca_scale = T,
      normalize = T,
      eta = 200.0,
      exaggeration_factor = 12.0,
      num_threads = parallel::detectCores() - 2
      )
## Performing PCA
## Read the 1544 x 42 data matrix successfully!
## OpenMP is working. 6 threads.
## Using no_dims = 3, perplexity = 420.000000, and theta = 0.100000
## Computing input similarities...
## Building tree...
## Done in 10.13 seconds (sparsity = 0.932734)!
## Learning embedding...
## Iteration 50: error is 44.145871 (50 iterations in 15.87 seconds)
## Iteration 100: error is 44.145871 (50 iterations in 18.00 seconds)
## Iteration 150: error is 44.145871 (50 iterations in 20.99 seconds)
## Iteration 200: error is 44.145871 (50 iterations in 22.81 seconds)
## Iteration 250: error is 44.145871 (50 iterations in 25.42 seconds)
## Iteration 300: error is 1.193916 (50 iterations in 21.90 seconds)
## Iteration 350: error is 0.319184 (50 iterations in 11.58 seconds)
## Iteration 400: error is 0.241505 (50 iterations in 11.28 seconds)
## Iteration 450: error is 0.229975 (50 iterations in 10.10 seconds)
## Iteration 500: error is 0.226926 (50 iterations in 10.35 seconds)
## Iteration 550: error is 0.226181 (50 iterations in 10.55 seconds)
## Iteration 600: error is 0.225912 (50 iterations in 10.54 seconds)
## Iteration 650: error is 0.225838 (50 iterations in 10.27 seconds)
## Iteration 700: error is 0.225819 (50 iterations in 10.23 seconds)
## Iteration 750: error is 0.225798 (50 iterations in 10.05 seconds)
## Iteration 800: error is 0.225781 (50 iterations in 10.04 seconds)
## Iteration 850: error is 0.225778 (50 iterations in 10.17 seconds)
## Iteration 900: error is 0.225771 (50 iterations in 10.58 seconds)
## Iteration 950: error is 0.225767 (50 iterations in 10.08 seconds)
## Iteration 1000: error is 0.225760 (50 iterations in 9.78 seconds)
## Iteration 1050: error is 0.225743 (50 iterations in 10.27 seconds)
## Iteration 1100: error is 0.225722 (50 iterations in 10.02 seconds)
## Iteration 1150: error is 0.225723 (50 iterations in 10.27 seconds)
## Iteration 1200: error is 0.225719 (50 iterations in 9.85 seconds)
## Iteration 1250: error is 0.225719 (50 iterations in 9.81 seconds)
## Iteration 1300: error is 0.225725 (50 iterations in 10.28 seconds)
## Iteration 1350: error is 0.225728 (50 iterations in 10.11 seconds)
## Iteration 1400: error is 0.225729 (50 iterations in 10.30 seconds)
## Iteration 1450: error is 0.225728 (50 iterations in 10.00 seconds)
## Iteration 1500: error is 0.225730 (50 iterations in 10.32 seconds)
## Iteration 1550: error is 0.225729 (50 iterations in 10.34 seconds)
## Iteration 1600: error is 0.225730 (50 iterations in 9.85 seconds)
## Iteration 1650: error is 0.225734 (50 iterations in 10.03 seconds)
## Iteration 1700: error is 0.225733 (50 iterations in 10.53 seconds)
## Iteration 1750: error is 0.225732 (50 iterations in 9.77 seconds)
## Iteration 1800: error is 0.225733 (50 iterations in 10.20 seconds)
## Iteration 1850: error is 0.225730 (50 iterations in 9.96 seconds)
## Iteration 1900: error is 0.225729 (50 iterations in 10.14 seconds)
## Iteration 1950: error is 0.225730 (50 iterations in 10.35 seconds)
## Iteration 2000: error is 0.225735 (50 iterations in 9.97 seconds)
## Fitting performed in 472.95 seconds.
tsne_df %>% 
  ungroup() %>% 
  select(!where(is.character)) %>% 
  select(!c(which(apply(., 2, var)==0))) %>% 
  unique() -> a

tsne_out %>% 
  .$Y %>% 
  as_tibble() %>% 
  ggplot() +
    geom_point(aes(V1, V2, color = a$rich))

tsne_out %>% 
  .$Y %>% 
  as_tibble() %>% 
  plot_ly(
    title = "Sample title",
    x = .$V1,
    y = .$V2,
    z = .$V3,
    type = "scatter3d",
    mode = "markers",
    color = a$rich
  ) %>% 
  layout(title = "Experiment 29")
tsne_out %>% 
  .$Y %>% 
  as_tibble() %>% 
  plot_ly(
    title = "Sample title",
    x = .$V1,
    y = .$V2,
    z = .$V3,
    type = "scatter3d",
    mode = "markers",
    color = a$rich
  ) %>% 
  layout(title = "Experiment 29")

2.2 Clusterização Experimento 66

library(Rtsne)
set.seed(42)

tsne_df <- cdr %>% 
  # filter(!rich == "poor") %>%
  filter(time == "final") %>% 
  filter(cycle == "R0_R4") %>% 
  filter(expgroup == "66")

tsne_df %>% 
  summarise(across(everything(), ~ all(sum(is.na(.x))))) %>% 
  select(where(isTRUE))
## # A tibble: 1 x 2
## # Groups:   expgroup, cycle [1]
##   expgroup cycle
##   <chr>    <chr>
## 1 66       R0_R4
tsne_df %>% 
  filter(is.na(threshold)) %>% 
  select(rich, threshold, quantity)
## # A tibble: 0 x 6
## # Groups:   expgroup, cycle, time [0]
## # … with 6 variables: expgroup <chr>, cycle <chr>, time <chr>, rich <fct>,
## #   threshold <dbl>, quantity <int>
set.seed(42)
tsne_out <- 
  tsne_df %>% 
    ungroup() %>% 
    select(!where(is.character)) %>% 
    select(!c(which(apply(., 2, var)==0))) %>% 
    unique() %>% 
    Rtsne(
      X = .,
      dims = 3,
      perplexity = 30,
      theta = 0.5,
      max_iter = 1E3,
      verbose = T,
      pca_center = T,
      pca_scale = T,
      normalize = T,
      eta = 200.0,
      exaggeration_factor = 12.0,
      num_threads = parallel::detectCores() - 2
      )
## Performing PCA
## Read the 8436 x 42 data matrix successfully!
## OpenMP is working. 6 threads.
## Using no_dims = 3, perplexity = 30.000000, and theta = 0.500000
## Computing input similarities...
## Building tree...
## Done in 14.15 seconds (sparsity = 0.015493)!
## Learning embedding...
## Iteration 50: error is 95.444063 (50 iterations in 25.25 seconds)
## Iteration 100: error is 92.761432 (50 iterations in 37.79 seconds)
## Iteration 150: error is 91.017719 (50 iterations in 19.85 seconds)
## Iteration 200: error is 90.890341 (50 iterations in 16.90 seconds)
## Iteration 250: error is 90.831025 (50 iterations in 17.12 seconds)
## Iteration 300: error is 2.973973 (50 iterations in 13.09 seconds)
## Iteration 350: error is 2.540478 (50 iterations in 12.58 seconds)
## Iteration 400: error is 2.310407 (50 iterations in 13.03 seconds)
## Iteration 450: error is 2.166549 (50 iterations in 13.19 seconds)
## Iteration 500: error is 2.067423 (50 iterations in 13.28 seconds)
## Iteration 550: error is 1.995633 (50 iterations in 13.70 seconds)
## Iteration 600: error is 1.942256 (50 iterations in 13.50 seconds)
## Iteration 650: error is 1.902282 (50 iterations in 13.35 seconds)
## Iteration 700: error is 1.871328 (50 iterations in 12.91 seconds)
## Iteration 750: error is 1.847140 (50 iterations in 13.84 seconds)
## Iteration 800: error is 1.828978 (50 iterations in 13.87 seconds)
## Iteration 850: error is 1.815785 (50 iterations in 14.25 seconds)
## Iteration 900: error is 1.805099 (50 iterations in 13.57 seconds)
## Iteration 950: error is 1.796396 (50 iterations in 14.03 seconds)
## Iteration 1000: error is 1.788672 (50 iterations in 14.26 seconds)
## Fitting performed in 319.36 seconds.
tsne_df %>% 
  ungroup() %>% 
  select(!where(is.character)) %>% 
  select(!c(which(apply(., 2, var)==0))) %>% 
  unique() -> a

tsne_out %>% 
  .$Y %>% 
  as_tibble() %>% 
  ggplot() +
    geom_point(aes(V1, V2, color = a$rich))

tsne_out %>% 
  .$Y %>% 
  as_tibble() %>% 
  plot_ly(
    x = .$V1,
    y = .$V2,
    z = .$V3,
    type = "scatter3d",
    mode = "markers",
    color = a$rich
  ) %>% 
  layout(title = "Experiment 66")